In recent years, advances in semiconductor integrated circuits for wireless and RF technology have dramatically changed our perceptions, use, and reliance upon portable electronic devices. The uses of wireless technology are widespread, increasing, and include, but are not limited to, telephony, Internet e-mail, Internet web browsers, global positioning, photography, diary, address book, and in-store navigation. Additionally, devices incorporating wireless technology have expanded to include not only cellular telephones, but Personal Digital Assistant (PDAs), laptop computers, palmtop computers, gaming consoles, printers, telephone headsets, portable music players, point of sale terminals, global positioning systems, inventory control systems, and even vending machines. Today, many of these devices are high volume consumer commodities supplied by businesses competing through integrated features and rapidly changing branding whilst reinforcing customers desire for small size, long battery life, and lightweight devices with increased roaming capabilities and download speeds. Within a matter of a few years, these systems have evolved from bulky cellular telephones offering voice and simple text messaging to lightweight compact multi-media players providing streaming live video and music alongside telephony, PDA functionality, integrated mega pixel CCD camera, and supporting BlueTooth™ wireless peripheral interfaces for headphones, microphones etc.
Semiconductor integrated circuits have been an important enabler of this rapid evolution by offering high volume, low cost solutions with low power consumption, small footprint and reduced component count when compared to discrete or hybrid solutions. There is significant economic and business pressure to ensure that these trends continue, whilst providing increased benefits to the manufacturers including reducing the number of chips for a chip set, providing multiple standards from a single chip set, and providing specifications and margin allowing today's chip sets to support the evolving bandwidth and spectral aspects of these systems and standards.
Indeed, such wireless semiconductor circuits today address a plethora of standards including, but not limited to, IEEE 802.11 WiFi, IEEE 802.16 WiMAX, quad-band GSM, EDGE, GPRS, and Global Positioning Systems. In many instances, such as IEEE 802.16e WiMAX with targeted data rates of 10 Mb/s at a 10 km range from a base station, the systems are also stretching the limits of performance in respect of transmitted power, received power, dynamic range, interference, efficiency and bandwidth.
All of these performance aspects impact the design and implementation of the RF amplifiers that form a critical element in both the transmitter and receiver portions of the transceivers within these devices. In order to increase efficiency wireless amplifiers have tended to move from the high linearity amplifier design typified by classes A, AB, and B, (http://en.wikipedia.org/wiki/amplifier) to non-linear amplifier classes such as C, D, E and F, where efficiencies over 90% can be achieved.
The evolution from low efficiency, high linearity amplifiers to very high efficiency non-linear amplifiers has circuit designers exploiting classical techniques to correct the distortion generated in the amplifier. For those skilled in the art such classical techniques fall into 5 different classes; feed-forward, feedback, pre-distortion, adaptive bias, and synthesis. However, the basic objective on any amplifier in the context of a wireless information transmission system is to provide an exact copy of the signal intended to be transmitted at the correct power level with highest possible power efficiency.
Feed-Forward: This approach employs an error signal that is extracted from an amplifier, commonly referred to as the power amplifier (PA) on the transmit path. Considering a PA for the following discussions, then the PA is corrected by subtracting a scaled (attenuated) version of the output signal of the PA from the input signal. If these signals are properly scaled, then the resulting signal contains only error information that is spectral energy generated by the non-linearity of the PA, noise from the PA, and energy resulting from non-flat frequency response of the PA, and none of the original signal. This error signal is then amplified appropriately, and subtracted from the output signal of the PA. Typically, the output signal from the amplifier is time delayed to account for the delay in the components of the error signal generation path. An example of the feed-forward approach is disclosed by Chen et al in “Article comprising a Power Amplifier with Feed Forward Linearizer using a RLS Parameter Tracking Algorithm” (U.S. Pat. No. 5,963,091) and described in respect of FIG. 1 subsequently. If the amplitude and phase are correct, feeding forward and subtracting the error removes all the error (distortion) generated in the PA in question and is powerful in that the approach corrects any error generated by the PA. However, it is a very inefficient correction technique as firstly, the amplifier boosting the error signal must itself be fairly large and linear as its output signal is generally combined with the PA output signal using a low ratio coupler to avoid losses on the main transmit path. Secondly, the time delay applied to the output signal of the PA due to the circuit delay in the error signal path can add significant loss.
Pre-Distortion: This correction approach employs a model of the PA to predict the distortion that will be generated in the amplifier being corrected. The modeled distortion is then added to the input signal provided to the amplifier, with appropriate phase adjustment, such that it cancels the distortion produced. The model can take many forms including, but not limited to, a look-up table of amplitude modulation (AM) and phase modulation (PM) transfer curves such as AM-AM and AM-PM curves, it can be non-linear electronic hardware, it can be DSP algorithms, or it can be a scaled model of the amplifier being corrected. An example of pre-distortion applied to an amplifier is presented by Midya et al “Scalar Cost Function based Pre-Distortion Device, Method, Phone and Base Station” (U.S. Pat. No. 6,240,278), as presented and discussed in respect of FIG. 2.
Pre-distortion is an efficient technique, in that there are no lossy circuit elements after the power amplifier, and there are no additional microwave circuit blocks that are inherently power-hungry. Further, the ability to use digital hardware, which has dramatically improved in capability in recent years, has made this a favored solution. However, predistortion can only easily cope with simple memoryless deterministic distortion and typically assumes that the AM-AM and AM-PM curves are static and do not depend on earlier events, operational conditions, or frequency of operation. Furthermore, the nature of the model used to predict the error must suit the amplifier. That is, the model must employ an appropriate order of non-linearity, or an appropriate number of entries in the look-up table.
This makes the pre-distortion system somewhat specific to the amplifier being corrected. Although digital hardware tends to be relatively low cost, a digital pre-distorter can be quite elaborate, requiring wide data bus and fast sample rates (usually at least 5× the Nyquist rate of the data rate within the uncorrected signal). Finally, the algorithms for adjusting the non-linearity (the weights on the non-linear work function or the elements in the look-up table) are complex and prone to finding local minima. Despite these limitations and disadvantages, pre-distortion has a significant share of the distortion correction solutions implemented today.
Adaptive Bias: In contrast to the previous approaches, adaptive bias technique does not attempt to correct distortion products but rather can improve either the linearity or minimize the distortion of the PA. In an adaptive bias system, the bias voltages on the terminals of the active device are adjusted to suit the instantaneous signal being transmitted. For example, the collector or drain voltage of a transistor amplifier can be increased during peaks in the amplitude of the input signal. This technique can be a simple way to make modest improvements to amplifier linearity, however it also is amplifier dependant, and large improvements in linearity are difficult to achieve. As a result, the adaptive bias approach has limited benefit to the very high efficiency but highly non-linear amplifier classes which suit the demands for low power consumption in wireless handheld devices.
Synthesis: This approach is more a general category of “synthesis techniques”, in which a linear PA is not used, but instead the signal at the output port of the amplification system is generated by combining 2 or more signals, each of the initial signal components not resembling the final signal being generated. Examples of this technique include Envelope Elimination and Restoration (EE&R), as discussed by Midya, Khan et al “Method, Device, Phone and Base Station for Providing Envelope-Following for Variable Envelope Radio Frequency Signals” (U.S. Pat. No. 6,141,541), and Linear Amplification with Nonlinear Components (LINC), as discussed by Okubo et al “Constant-Amplitude Wave Combination Type Amplifier” (U.S. Pat. No. 5,287,069).
In EE&R, a constant amplitude signal with variable phase is amplified, and the envelope restored by varying the collector or drain voltage of the transistor amplifier. In LINC, two constant amplitude signals are combined in various phases to generate a signal with the correct amplitude and phase. These techniques have specific applications, but all have significant drawbacks in respect of power, bandwidth, efficiency and linearity.
Feedback: The general technique of feedback correction goes back nearly 80 years, see for example H. S. Black “Wave Translation System” (U.S. Pat. No. 2,102,671; filed 1932). In particular, negative feedback tends to act to reduce variability of gain, and reduces distortion introduced in an amplifier. The actual feedback may be implemented in many different forms. Perhaps the simplest form of feedback being referred to as circuit-level feedback, wherein an electrical linkage couples some of the energy from the output port of an amplifier back to the input port. Considering a single transistor amplifier, such approaches include shunt-feedback, where a resistor is placed between the drain and gate of a transistor, and series feedback, where an inductor is inserted into the source of a transistor.
Circuit level feedback can be applied to multiple stages, but the signal delay through the stages must be accounted for. If there is too much delay, then the negative feedback will, at some frequency, become positive feedback, and an oscillation results because a portion of the output signal (at a particular set of frequencies) adds constructively to the input signal. With each passage through the loop, the signal increases to the point where all energy can be found in those particular frequencies where the combination is most constructive. Of course, as long as the gain is less than unity at the frequency at which the phase shift through the feedback loop is 360 degrees, the amplifier will remain stable. While simple circuit level feedback is widely used it suffers from one major fault in that it necessarily decreases the gain of the amplifier. Hence, higher levels of correction necessarily increasing gain which is an issue at RF frequencies where gain is difficult and expensive to achieve.
As a result, other feedback approaches have been established to act upon only the information on the envelope of the signal being amplified, which has the advantages that the RF gain of the amplifier is undiminished, and the feedback circuit may be implemented in baseband. Two such approaches being Cartesian Feedback and Polar Feedback. Considering, Cartesian Feedback, a typical approach is presented by Leitch “Gain/Phase Compensation for Linear Amplified Feedback Loop” (U.S. Pat. No. 4,933,986), in demodulating the signal at the output port of the amplifier into in-phase (I) and out-of-phase (Q) components in the Cartesian coordinate system. These I- and Q-output signal components can be compared to the I- and Q-input signals, and the results applied to an I/Q modulator at the input port of the amplifier to provide correction.
An advantage of the Cartesian feedback approach is that the filtering that keeps the amplifier stable can now be done at baseband, which is advantageous, as it no longer needs to be tuned with the RF frequency of operation of the amplifier. The design of this filter is critical, however, as the filter's frequency response is superimposed onto the gain of the amplifier. Further, delay through the amplifier is critical, as it determines the bandwidth of the filter required to stabilize the feedback loop, and therefore the instantaneous bandwidth of the system.
In contrast, Polar feedback resolves the amplitude and phase elements of the output signal, compares these to the amplitude and phase characteristics of the input signal, and uses the resulting error terms to control a polar modulator placed before the input port of the power amplifier, thereby closing the loop. This approach is presented subsequently in FIG. 4. Both Cartesian and Polar feedback techniques suffer from the limitations of the loop filter that is used to ensure that the gain of the loop rolls off fast enough with increasing frequency such that the loop remains stable. In the prior art, this loop filter adds delay and amplitude variation with frequency across the band.
It is possible, however, to configure feedback loops in which just the error is fed back, such an approach being presented by Huang “Wideband, Phase Compensated Amplifier with Negative Feedback of Distortion Components in the Output Signal” (U.S. Pat. No. 4,276,514). This approach generates the error term by subtracting the input signal from the attenuated version of the output signal. This error term is fed back into the input port of the amplifier in anti-phase (inverted) to effect the correction. This is advantageous, as the frequency response of the filter is now interposed only on the error signal and the main signal is not at all affected. Ripple in the frequency response of this filter will vary the amount of correction applied, but not the gain of the main signal. Implementations of error feedback have been limited, primarily as initial publications on error feedback, such as John McRory “An RF Amplifier for Low Intermodulation Distortion” (1994 MTT-S Digest, pp. 1741-) resulted in minimal improvements being observed. Two significant challenges in implementing this architecture exist. Firstly, the implementation of the tunable filter which must be at RF, but have a finesse adequately high to maintain stability, and secondly in designing the components in the loop to have minimal delay so that a useful correction bandwidth results.
The classical techniques presented supra all suffer from one or more of the following impairments:                Poor Efficiency        Limited Bandwidth        Complexity        Limited Effectiveness        
It would be apparent that such limitations result in system designers and circuit designers trading aspects of performance and cost in implementing commercial systems with non-linear amplifiers for improved efficiency using these prior art approaches.
It would therefore be advantageous to provide a linearization solution for an RF amplifier that addresses these drawbacks of prior art approaches whilst leveraging increased integration potential within semiconductor integrated circuits for lowering cost, footprint and power consumption. It would be particularly beneficial if the linearization solution addressed the increasing fractional bandwidth of today's increasing data rate wireless protocols.